Want more qualified Shopify traffic from AI search?
Run a free Naridon scan to see which prompts, products, and AI engines can send more ready-to-buy visitors.
TL;DR: ACP, AP2, and MCP are not rivals, they are different layers of one agentic commerce stack. MCP (Anthropic) is the tool and context layer, what an agent can read and call. A2A is the agent to agent layer, how agents talk to each other. ACP (OpenAI and Stripe) is the checkout layer, how an agent buys from a merchant. AP2 (Google) is the payments and authorization layer, how the buyer's intent and the merchant's charge are cryptographically authorized. They are designed to compose, so a single purchase can use several at once. Because no single standard has clearly won, the durable merchant move is to invest in the layer-independent fundamentals every protocol depends on: an accurate, structured catalog and strong AI search visibility.
If you search for "ACP vs AP2 vs MCP" you get a lot of hot takes about which protocol will win agentic commerce. That framing is wrong. These are not competing products fighting for one slot. They are layers of a stack, each solving a different part of the problem of letting an AI agent shop and pay on a person's behalf. This is the fair map: what each layer does, who is behind it, its status as of mid 2026, and what actually matters if you run a store. We build a Shopify AI visibility app, so we have a stake in one narrow part of this, and we have kept the rest of the map neutral so it is genuinely useful.
The agentic commerce stack at a glance
Here is the whole landscape in one table. Read it top to bottom as roughly context, then discovery and negotiation, then checkout, then payment. Status is hedged where a standard is young; verify current versions against the official sources linked below before you build.
| Protocol | Who is behind it | Layer | What it does | Status (mid 2026) |
|---|---|---|---|---|
| MCP (Model Context Protocol) | Anthropic (open) | Tool / context | Defines how a single agent reads data and calls tools and APIs. It is how an agent reaches the other layers, not a payments protocol. | Open, widely adopted since late 2024 |
| A2A (Agent2Agent) | Google, now a Linux Foundation project | Agent to agent | Lets agents discover and talk to each other, for example a shopping agent negotiating with a merchant agent. | Open, donated to the Linux Foundation in 2025 |
| ACP (Agentic Commerce Protocol) | OpenAI and Stripe | Checkout / buy | The checkout flow between a buyer's agent and a merchant: create session, update shipping and variants, complete purchase, cancel. | Open (Apache 2.0), beta, spec versioned 2026-04-17 |
| AP2 (Agent Payments Protocol) | Google, 60+ launch partners | Payments / authorization | Authorizes the buyer's intent and the merchant's charge using signed Mandates (Intent, Cart, Payment) carried as W3C Verifiable Credentials. | Open, announced Sept 2025, stablecoins first class |
| UCP (Universal Commerce Protocol) | Google, Shopify, and a retail coalition | Catalog / cart | Standard schemas for an agent to discover products, build a cart, and track orders. Mention lightly, still emerging. | Emerging (2026), verify before building |
| x402 | Coinbase | Crypto micropayments | Instant stablecoin and crypto payments over HTTP, reviving the HTTP 402 "Payment Required" status code. | Live for crypto-native and machine to machine payments |
The four core layers, one at a time
MCP: what an agent can read and call
MCP, the Model Context Protocol, came from Anthropic and is open. It is the plumbing beneath everything else. MCP standardizes how a single agent connects to data sources and calls tools and APIs, so a model can pull in a product catalog, query inventory, or trigger an action without a bespoke integration each time. It is not a payments or checkout protocol. Think of it as how the agent reaches the commerce layers in the first place. If you want the primary source, see modelcontextprotocol.io.
A2A: agents talking to each other
A2A, Agent2Agent, was launched by Google and has since been donated to the Linux Foundation with broad industry backing. Where MCP connects one agent to tools, A2A connects agents to other agents. In a commerce setting that means a buyer's shopping agent could discover and negotiate with a merchant's agent, comparing options and coordinating a task across systems. It is the interoperability layer that keeps agents from being walled gardens.
ACP: how an agent buys
ACP, the Agentic Commerce Protocol, is co-developed and maintained by OpenAI and Stripe. It is the checkout layer. ACP is open sourced under Apache 2.0, uses date-based versioning (the spec is tagged 2026-04-17 and in beta), and ships as two OpenAPI specs: one for the checkout flow and one for payment delegation. The checkout flow is a small set of REST endpoints the agent calls on the merchant: create a checkout session, update it with shipping and variant choices, read its state, complete the purchase, or cancel.
Payment in ACP runs on Stripe's Shared Payment Token, a primitive that lets an app like ChatGPT initiate a charge scoped to a specific amount and merchant without exposing the buyer's card credentials. Crucially, the merchant stays the merchant of record: they accept or decline the order, charge through their existing payment provider, and own fulfillment, returns, and the customer relationship. ACP does not replace the store's backend. The spec lives at github.com/agentic-commerce-protocol.
AP2: how intent and charge are authorized
AP2, the Agent Payments Protocol, is Google's answer to a different question: not "how does the agent buy" but "how do we prove the human authorized this." Announced in September 2025 with more than 60 launch partners including Mastercard, PayPal, Coinbase, American Express, and Salesforce, AP2 introduces three signed Mandates, Intent, Cart, and Payment, carried as W3C Verifiable Credentials. Together they create a cryptographic chain of evidence that the buyer intended this purchase and authorized this exact charge, which matters a lot when a machine is spending your money. AP2 also treats stablecoins as a first-class payment method alongside cards and bank transfers. See ap2-protocol.org and the Google Cloud announcement.
The two you will also hear about: UCP and x402
Two more names show up in these conversations, and both are worth knowing lightly. UCP, the Universal Commerce Protocol, is an emerging effort associated with Google, Shopify, and a coalition of retailers to standardize the schemas an agent uses to discover products, build a cart, and track orders. It is early, so verify the details before you plan around it. x402, from Coinbase, is a payment protocol built on the long-dormant HTTP 402 status code, aimed at instant stablecoin and crypto micropayments between machines. Neither is a Shopify merchant's day-one concern, but both reinforce the same point: the stack is filling in layer by layer, not converging on one standard.
ACP vs AP2: the comparison people actually search for
Because the names are so close, "difference between ACP and AP2" and "agent payments protocol vs agentic commerce protocol" are the two queries people get stuck on. The clean answer: they operate at adjacent layers and can work in the same transaction. ACP (OpenAI and Stripe) defines the conversation between a buyer's agent and a merchant to assemble and complete an order. AP2 (Google) defines how that order's intent and charge get signed and trusted so any payment method, including stablecoins, can settle with proof the human agreed. One is the checkout handshake, the other is the authorization envelope around the money. They are not substitutes.
Do they compete? No, they compose
The reusable mental model is short. MCP is what an agent can read and call. A2A is agents talking to each other. ACP is how an agent buys. AP2 is how the buyer's intent and the merchant's charge are authorized. A single real-world purchase might use MCP to read your catalog, A2A for the agents to coordinate, ACP for the checkout flow, and AP2 to authorize the payment. That is why "which one wins" is the wrong question. Treat them as camps in a stack, not gladiators.
It is also why the early hype got ahead of reality. The "one-click buy inside ChatGPT" story stumbled: OpenAI narrowed its Instant Checkout rollout in early 2026 after accurate product data, multi-item carts, and onboarding at scale proved hard. Products still surface inside AI assistants, but many purchases still complete on the merchant's own storefront. The protocols and the broader shift are real and strategic; the timeline is messier than the launch posts suggested.
What a Shopify merchant should actually do
Here is the practical takeaway, and it is deliberately boring. You cannot safely bet your store on one protocol this early. The standards are still stabilizing, different buying agents will adopt different combinations, and the checkout layer that looks dominant today may not be the one that matters in a year. So do not build your strategy on ACP versus AP2. Build it on the layer every one of these protocols quietly depends on.
Agents can only buy what they can first find and trust. Every protocol above assumes your catalog is accurate, structured, and machine-readable, and that the AI engines shoppers ask can actually see and recommend your products. That layer is protocol-independent. Whether the winning checkout standard turns out to be ACP, AP2, UCP, or something not yet announced, an accurate structured catalog plus strong Generative Engine Optimization pays off under all of them. That is the safe bet.
This is the narrow slice we work on. Naridon is a Shopify-native app that tracks whether AI engines cite and recommend your products across ChatGPT, Perplexity, Google AI Overviews, Gemini, and Copilot, and its Autopilot applies the fixes that make a catalog machine-readable (structured data and JSON-LD, product copy, llms.txt), with every change revertible. It is not a payments processor and it does not implement ACP or AP2 checkout for you. It sits upstream of them, as the visibility and catalog-readiness layer, and it is free to start ($0), then $49 a month if you need more. For the merchant playbook, see our Agentic Commerce Protocol guide for Shopify and how to prepare your Shopify store for agentic commerce, or compare plans on the pricing page.
The protocol wars will keep making headlines. The merchants who win the agentic era will be the ones whose products every agent can already find, trust, and transact, no matter which layer of the stack ends up on top.
Frequently asked
- What is the difference between ACP and AP2?
- They sit at different layers of the same stack. ACP (Agentic Commerce Protocol, from OpenAI and Stripe) is the checkout layer: it defines how a buyer's agent creates a checkout session with a merchant, updates shipping and variants, and completes the purchase. AP2 (Agent Payments Protocol, from Google) is the payments and authorization layer: it uses signed Mandates carried as W3C Verifiable Credentials to prove that the buyer authorized this intent and this specific charge. In short, ACP is how an agent buys and AP2 is how the intent and the charge are cryptographically authorized.
- What is the difference between MCP, ACP, and AP2?
- MCP (Model Context Protocol, from Anthropic) is the tool and context layer: it defines how a single agent reads data and calls tools and APIs. It is how an agent reaches the other layers, not a payments protocol. ACP is the checkout layer (how an agent buys from a merchant). AP2 is the payments and authorization layer (how the buyer's intent and the merchant's charge are signed and trusted). They are designed to compose, and a single buying flow can use all three.
- Is AP2 the same as the Agentic Commerce Protocol?
- No. AP2 is the Agent Payments Protocol from Google, focused on authorizing payments. ACP is the Agentic Commerce Protocol from OpenAI and Stripe, focused on the checkout flow. The names are easy to confuse, but they solve different problems and can work together in the same transaction.
- Do ACP, AP2, and MCP compete with each other?
- Not directly. They are complementary layers of one agentic commerce stack rather than one winner takes all. MCP is what an agent can read and call, A2A is agents talking to each other, ACP is how an agent buys, and AP2 is how the buyer's intent and the merchant's charge are authorized. A single flow can use several at once, so the useful question is which layer a given standard addresses, not which one will kill the others.
- Which agentic commerce protocol should a Shopify merchant adopt?
- You cannot safely bet the store on one standard this early, because the layers are still stabilizing and different buying agents will use different combinations. The durable move is to invest in the layer-independent fundamentals that every protocol depends on: an accurate, structured, machine-readable catalog and strong AI search visibility, so any buying agent can find, trust, and transact your products whichever checkout or payment standard it uses.
- Are stablecoins part of these agentic payment protocols?
- Yes, in some of them. Google's AP2 treats stablecoins as a first-class payment method alongside cards and bank transfers. Separately, x402 from Coinbase is a crypto and stablecoin micropayment protocol built on the HTTP 402 status code. ACP, by contrast, runs on Stripe's Shared Payment Token and the merchant's existing payment provider.
Key concepts
Plain-language definitions of the terms in this guide.
Ready to rank for these conversations?
Join early adopters who are already capturing AI search traffic.